View
122
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Published in The 7th International Conference on Information Technology and Applications (ICITA 2011), Sydney Australia, 21 Nov - 24 Nov 2011 The purpose of this paper is to provide additional analysis of image quality and Henry Classification on Finger location on a single sensor. One hundred and sixty nine individuals provided six impressions of their left index, left middle, right index, and right middle fingers. The results show that there is significant difference in image quality, Henry classifications, and zoo animal distribution across the four finger locations under study. The results of this research show that location is an important consideration when developing enrollment best practices for single print systems.
Citation preview
Abstract--The purpose of this paper is to provide
additional analysis of image quality and Henry
Classification on Finger location on a single sensor. One
hundred and sixty nine individuals provided six impressions
of their left index, left middle, right index, and right middle
fingers. The results show that there is significant difference
in image quality, Henry classifications, and zoo animal
distribution across the four finger locations under study.
The results of this research show that location is an
important consideration when developing enrollment best
practices for single print systems.
Index Terms--Biometrics, Fingerprint Recognition,
Henry Classifications, Zoo Plots
I. INTRODUCTION
In 1888, Edward Henry developed a classification schema for
pattern types of fingerprints. There are three main types of
classification; loops, whorls, and arches [1]. Additional
subcategories can also be created, and for the purposes of this
paper, we use six classifications generated by a commercially
available Henry Classifier. These are left loop, right loop, plain
arch, tented arch, whorl and scar. This is a similar classification
approach as noted in [1]. As noted in [2] and [3], the distribution
of Henry classifications are not homogenous. For example, the
predominant Henry Classification in [3] for the left thumb and
left index is the right loop (51.26%, 34.61% of occurrence
respectively); the right thumb and right index is left loop
(58.44%, 39.00% of occurrence respectively). In a recent study,
the most frequent pattern was the ulnar loop, both in the general
population and the gender distribution [4]. The motivation of
this work is to add to the body of literature on the distribution of
these classifications, across different fingers, and to expand the
zoo-plot methodology proposed by [5].
M. Brockly is with the Technology, Leadership, and Innovation Department
of Purdue University, West Lafayette IN 47907 USA (telephone:
765-494-2311, e-mail: [email protected]).
S. Elliott, is with the Technology, Leadership, and Innovation Department
of Purdue University, West Lafayette IN 47907 USA (telephone:
765-494-2311, e-mail: [email protected]).
ISBN: 978-0-9803267-4-1
II. RESEARCH QUESTIONS AND MOTIVATION
The purpose of this paper to understand how the Henry
classification, image quality, minutiae count and performance
are all impacted by finger location. There are three motivating
factors – the first is to update a previous paper [2], that
examined a similar problem and used the same sensor. Our
paper differs in that a different population was used, and
different classification, quality and performance tools are
implemented. Furthermore, in that paper, performance was
shown by a series of detection error tradeoff curves, in this
paper, we are evaluating the performance of the specific finger
locations by looking at the zoo plot analysis – the second
motivation behind this paper. This is a relatively new approach
in visualizing data in biometric systems. A detection error
tradeoff curve typically used in the biometric literature presents
a graphical image of biometric performance, but only at a global
level. A zoo plot analysis will examine whether there are any
differences in a user’s genuine and impostor match scores
across the Henry Classification, Finger Location, and Image
Quality. In order to calculate a zoo plot, the methodology
outlined in [5] was applied. The zoo plot outputs four different
“animals” called chameleons, phantoms, doves, and worms.
Each of these animals has specific traits, or characteristics based
on their respective match scores. Subjects that are chameleons
are characterized by high match scores, and therefore they are
hardly cause false rejects, but are likely to cause false accepts.
Conversely, phantoms lead to low match scores, regardless of
who they match against. The other two animals are worms –
these animals are the worst users, and typically cause a
proportionately high numbers of errors. The best users are
worms, who have high matching scores as well as low impostor
scores. A frequency of each animal’s occurrence by
classification and location will be presented in the results
section. The third motivation is to add to the current body of
knowledge relating to the Henry Classification, and to gain a
better understand in the issues relating to finger location, so that
the results of this paper can be fed back into the development of
the Human Biometric Sensor Interaction model [6]. The
remainder of the paper covers the experimental methodology,
results, and finally conclusions and recommendations for future
work.
III. METHODOLOGY
A. Data Collection
This study used data collected from a larger study [7]. This
dataset was sub-sampled to one sensor, and only those subjects
that provided six successful impressions on the left index, left
Image Quality, Performance, and Classification – the
Impact of Finger Location
Michael E. Brockly, Stephen J. Elliott
The 7th International Conference on Information Technology and Applications (ICITA 2011)
300
middle, right index, and right middle were included. A total of
169 subjects are included in the analysis, the demographic
information is shown below in Table I:
Total Population Male Female N/A
169 118 49 2
Self-disclosed
information
Office Manual N/A
169 148 16 5
Table I. Demographic Information
The sensor used was the Identix® DFR-2080. Sensor
information with compliance to standard ISO/IEC JTC 1
N-29120 standard in Table II:
Manufacturer: Model: Dimensions:
Identix DFR-2080 76 x 38 x 83 mm
Resolution: Type: Platen size:
500 dpi Optical Touch 15 x 15 mm
Table II. Sensor Information
The breakdown of subjects by age is shown in Figure 1.
6154534947434038373635343331302928272625242322212019
40
30
20
10
0
Age
Fre
qu
en
cy
User distribution of age
Fig. 1. Distribution of Age
The variables of interest in this study were minutiae count,
quality score, Henry classification, and the metrics associated
with the zoo plot analysis. The minutiae count and image quality
was derived through using a commercially available tool
(Aware M1 pack version 3.0.0). The Henry classification was
extracted, and matching performance calculated using
Neurotechnology Megamatcher version 4.0.0. The zoo plots
were calculated using Performix with the output from the
matcher.
IV. RESULTS
The results section will answer the following research
questions: (1) how does the Henry classification change across
finger locations; (2) does image quality change across the finger
locations; (3) does minutiae count change across the finger
locations; (4) does finger location impact performance
(presented through the use of the zoo animal model developed
by [5]).
A. Henry Classification across Finger Locations
A total of 4,080 samples were processed, with the Henry
classifications across the finger locations noted in the table
below.
Henry LI LM RI RM
No % No % No % No %
Whorl 322 31.6 235 23.0 301 29.5 212 20.8
Left Slant
Loop
421 41.3 675 66.2 259 25.4 46 70.0
Right Slant
Loop
190 18.6 35 3.4 366 35.9 714 4.5
Tented
Arch
54 5.3 46 4.5 24 5.2 5 2.8
Plain Arch 32 3.1 24 2.4 39 3.8 14 1.4
Scar 1 0.1 5 0.5 2 0.2 5 0.5
Table III. Henry Classification by Finger Location
These observations are in line with previous studies that
examined the distribution of Henry Classifications. For example
in [4], the distribution of index fingers for whorls is 28.8%
(31.6% for this study on the left index; 29.5 on the right index).
In terms of ranking, loops for the index finger comprise of
43.4% which is approximately in line with the findings of this
study.
B. Image Quality
Image quality can impact the performance of a fingerprint
recognition system. Figure 2 shows interval plot of image
quality across the four finger locations.
Fig. 2. Image Quality vs. Finger Location
The results show that image quality differed between the
finger locations; one-way ANOVA F(3,4076) = 22.01, p<0.001.
C. Minutiae Count across Finger Locations
The process was repeated to establish whether there were any
significant differences in minutiae count across Henry
Classifications and finger location. The minutiae count differed
between Henry Classifications; one-way ANOVA F(5, 4074) =
94.47, p=<0.001, with the grouping information using the
Tukey HSD method shown below in Table IV.
Henry Class Grouping
Whorl A
301
Scar A B
Left Slant Loop B
Tented Arch B
Right Slant Loop B
Plain Arch C
Table IV. Henry Classification by Grouping
Fig. 3. Minutiae vs. Finger Location
Figure 3 shows an interval plot of the number of minutiae
across finger locations. Image quality differed between the
finger locations; one-way ANOVA F(3,4076) = 9.62, p<0.001.
The grouping information using the Tukey HSD is shown in
Table V. The means that do not share a letter are significantly
different.
Finger Location Grouping
Left Index A
Left Middle A
Right Middle B
Left Middle B
Table V. Finger Location by Grouping
D. Performance at Different Finger Locations
Figure 4 shows the detection error tradeoff curve for all finger
locations. This particular dataset had an Equal Error Rate of
0.30%. Figure 5 shows the zoo plot and the spread on both axis
indicating that there are a set of poorly performing users. This
can be seen in more detail in Figure 6, where the distribution of
animals by finger location is illustrated. There is a high
frequency of chameleons and doves – especially in the right
index location, followed by left index.
Fig. 4. Detection Error Tradeoff Curve
Fig. 5. Zoo Plot
Fig. 6. Distribution of Zoo Animals across Finger Location
302
V. CONCLUSIONS
We can see that finger location has a contributing factor to
performance (the zoo plots being an indicator of this). There are
also differences in minutiae counts across the four finger
locations. Future research should investigate whether these
results hold for other fingerprint sensors with the same
population – for example, are the Henry Classifications
consistent across different types of different sensors. In
addition, the zoo plots provide a deeper understanding of
performance, and therefore, understanding the characteristics of
these animals, and how poor performance, especially in the case
of worms and chameleons can be addressed. A deeper analysis
of the genesis of the poor performing animals is probably
justified.
REFERENCES
[1] J. Wayman, A. K. Jain, D. Maltoni, and D. Maio, “An Introduction to
Biometric Authentication Systems,” in Biometric Systems, J.
Wayman, A. Jain, D. Maltoni, and D. Maio, Eds. London: Springer,
2005, pp. 1-20.
[2] M. R. Young and S. J. Elliott, “Image Quality and Performance
Based on Henry Classification and Finger Location,” in 2007 IEEE
Workshop on Automatic Identification Advanced Technologies,
2007, pp. 51-56.
[3] H. Jarosz, J. C. Fondeur, and X. Dupré, “Large-scale identification
system design,” in Biometric Systems: Technology, Design and
Performance Evaluation, London: Springer-Verlag, 2005, pp.
263–287.
[4] M. D. Nithin, B. M. Balaraj, B. Manjunatha, and C. M. Shashidhar,
“Study of fingerprint classification and their gender distribution
among South Indian population,” Journal of Forensic and Legal
Medicine, vol. 16, pp. 460-463, 2009.
[5] N. Yager and T. Dunstone, “The biometric menagerie.,” IEEE
transactions on pattern analysis and machine intelligence, vol. 32,
no. 2, pp. 220-30, Mar. 2010.
[6] E. P. Kukula, M. J. Sutton, and S. J. Elliott, “The
Human–Biometric-Sensor Interaction Evaluation Method: Biometric
Performance and Usability Measurements,” IEEE Transactions on
Instrumentation and Measurement, vol. 59, no. 4, pp. 784-791, Apr.
2010.
[7] S. Modi, “Analysis of Fingerprint Sensor Interoperability on System
Performance,” Purdue University, 2008.
303